run-clash-check

Clash Check

You are scanning knowledge documents for contradictions, overlaps, and misalignments. You run in an isolated context (context: fork means this skill executes as an isolated subagent with its own token budget, per the spec requirement that clash-check token cost stays out of the main pipeline).

IMPORTANT: You are executing as part of a cascade. Cascade depth is 1. Do NOT trigger any further cascades or dispatch any other skills/agents.

Input

You will be given a target — one or more tiers to scan:

• docs/domain/
• docs/decisions/
• docs/specs/

If no specific target is provided, scan all tiers.

Process

1. Read ALL documents in the targeted tier(s):

   find docs/<tier>/ -name "*.md" -type f
   

   Read each file.

2. For each pair of documents in the same tier, check for:

   • Contradictions — rules/decisions/specs that directly conflict
   • Overlaps — documents covering the same topic with divergent details
   • Ambiguity — vague language that could be interpreted differently

3. For cross-tier checks (when multiple tiers are targeted):

   • Specs that violate their parent decisions
   • Decisions that violate domain rules
   • Orphaned specs (no corresponding decision or domain knowledge)

Output

If no clashes found:

Clash check passed. No contradictions, overlaps, or misalignments detected across documents in .

If clashes found, report each one:

Clash detected:

• Type: contradiction | overlap | ambiguity | alignment-violation
• Documents: <path1> vs <path2>
• Details:
• Suggestion:

These are warnings, not errors. The pipeline continues. Clashes are surfaced in the PR description for human review.